The 90/10 phenomenon in directed signed social networks

TL;DR

This paper investigates the sign labeling patterns in directed signed social networks, revealing a consistent 90/10 distribution of friend and foe labels and analyzing how randomness influences human labeling behavior.

Contribution

It introduces an empirical analysis of sign distribution in social networks, highlighting the coexistence of randomness and non-randomness in human labeling behavior.

Findings

90/10 sign labeling phenomenon observed

Entropy $S_{out}$ is reduced by non-random labeling

Coexistence of randomness and non-randomness in human behavior

Abstract

We empirical study the signs' property in the directed signed social networks of Slashdot and Epinions by using an reshuffled approach. Through calculating the entropy $S_{out}$ and the giant component $G$, we find an interesting 90/10 phenomenon: each individual labels his/her neighbors as friends with $90\%$ or foes with $10\%$ uniformly random from the macroscopic perspective. We also find that the entropy $S_{out}$ is suppressed by the non-randomness of labeling sign. Our present work can prove how do the randomness and the non-randomness coexist in human behavior of labeling signs, qualitatively.